Mask-RCNN detection of COVID-19 pneumonia symptoms by employing Stacked Autoencoders in deep unsupervised learning on Low-Dose High Resolution CT
This paper applies AI (artificial intelligence) technology to analyze low-dose HRCT (High-resolution chest radiography) data in an attempt to detect COVID-19 pneumonia symptoms. A new model structure is proposed with segmentation of anatomical structures on DNNs-based (deep learning neural network) methods, relying on an abundance of labeled data for proper training. The model improves the existing techniques used for low-dose HRCT image inspection through an application of Stacked Autoencoders (SAEs) structures using the segmentation function for the area object detection model on Mask-RCNN. As a result, the proposed approach can quickly analyze X-ray images in detecting abnormalities in patients with lab-confirmed coronavirus even before clinical symptoms appear. In addition to detecting early abnormalities, area object detection model reveals a finding not seen in the latest cases of COVID-19. Most noteworthy, the study has shown that all COVID-19 patients exhibit an associated bilateral pleural effusion. The features are augmented to the model for the improvement of detection quality improvement and the shorten of the examination period.